Spectral prediction of all dielectric nanopore metasurface based on DBO-DNN model

Abstract

Based on the optical properties of symmetric structures independent of each other in the orthogonal direction, a class of all-dielectric nanohole array metasurfaces symmetrical along the diagonal is designed. By adding nanopores of different shapes to break the symmetry of the periodic unit structure, the double Fano resonance is excited. The spectral characteristics of metasurfaces with the same structure type are studied by finitedifference timedomain (FDTD) method. The deep neural network (DNN) is used to establish the nonlinear mapping relationship between the input structural parameters and the transmission spectrum. The number of hidden layers in the DNN and the number of neurons in each layer are optimized by the dung beetle optimization (DBO) algorithm. Therefore, the number of hidden layers of the model is determined to be 5, and the number of neurons in each layer is 120, 30, 150, 60, 90, respectively. The mean square error (MSE) is used to evaluate the training effect of DNN after optimization search. After 35,000 epochs of training, MSE is reduced to 0.0003926. The influence of different gradient descent optimization algorithms on the prediction results is explored respectively, and it is found that Adamax is the most effective. The results show that the prediction model can predict the spectrum within 1 s. Compared with the traditional simulation method, the simulation time is effectively saved. Meet the requirements of efficient and rapid design of ultra-thin lenses. For the same type of metasurface structure, the transmission spectrum can be accurately predicted without multiple data sets.